Around 10ms for trivial reads in Postgres feels a bit slow to me, if the amount of data in the blob is small and the database (or just this table) fits into RAM. Not sure where exactly the overhead is in this case, but I wouldn't attribute it to Postgres alone without further investigation.